US7167942B1

US7167942B1 - Dynamic random access memory controller

Info

Publication number: US7167942B1
Application number: US10/458,049
Authority: US
Inventors: Nafea Bishara
Original assignee: Marvell International Ltd
Current assignee: Marvell Asia Pte Ltd; Marvell Semiconductor Inc
Priority date: 2003-06-09
Filing date: 2003-06-09
Publication date: 2007-01-23
Also published as: US9430379B1

Abstract

An apparatus, and method and computer program thereof, comprises a plurality of ports each adapted to receive packets of data; a memory controller core adapted to generate one or more memory transactions for each of the packets of the data, wherein each memory transaction comprises a payload having a size of m bytes, and wherein the payloads contain the data; a memory comprising a plurality of memory banks adapted to store the data, wherein the memory can receive no more than n bytes of data in a single memory transaction; and a memory interface adapted to transmit the memory transactions to the memory; wherein m=kn and k is an integer.

Description

BACKGROUND

The present invention relates generally to memory control. More particularly, the present invention relates to controlling dynamic random access memory having multiple memory banks.

All processors, such as those found in routers and switches, rely on some form of memory for data storage. The two types most commonly employed are static random access memory (SRAM) and dynamic random access memory (DRAM). Because DRAM is cheaper and physically smaller than SRAM, it is employed whenever possible.

SUMMARY

In general, in one aspect, the invention features an apparatus comprising a plurality of ports each adapted to receive packets of data; a memory controller core adapted to generate one or more memory transactions for each of the packets of the data, wherein each memory transaction comprises a payload having a size of m bytes, and wherein the payloads contain the data; a memory comprising a plurality of memory banks adapted to store the data, wherein the memory can receive no more than n bytes of data in a single memory transaction; and a memory interface adapted to transmit the memory transactions to the memory; wherein m=kn and k is an integer.

Particular implementations can include one or more of the following features. k=1. The memory controller core is further adapted to direct each one of the memory transactions corresponding to one of the packets to a different one of the memory banks than the one of the memory banks to which the memory controller core directed the previous one of the memory transactions corresponding to the one of the packets. The memory banks define a plurality of buffers, and wherein the memory controller core is further adapted to direct each one of the memory transactions corresponding to one of the packets to the same buffer of the memory until all of the memory banks at that buffer contain data from the one of the packets. One implementation features a network switch comprising the apparatus. Another implementation features a traffic manager comprising the apparatus. Another implementation features a multi-port media access controller comprising the apparatus.

In general, in one aspect, the invention features a memory controller comprising at least one agent interface, wherein each agent interface is adapted to receive data from a respective memory agent; a memory controller core adapted to generate one or more memory transactions, wherein each memory transaction comprises a payload having a size of m bytes, and wherein the payloads contain the data; and a memory interface adapted to transmit the memory transactions to a memory, wherein the memory comprises a plurality of memory banks and can receive no more than n bytes of the data in a single memory transaction; wherein m=kn and k is an integer.

Particular implementations can include one or more of the following features. k=1. The memory controller core is further adapted to direct each one of the memory transactions to a different one of the memory banks than the one of the memory banks to which the memory controller core directed the previous one of the memory transactions. One implementation features a network switch comprising the memory controller. Another implementation features a traffic manager comprising the memory controller. Another implementation features a multi-port media access controller comprising the memory controller.

In general, in one aspect, the invention features a method and computer program for transmitting data to a memory comprising a plurality of memory banks, wherein the memory can receive no more than n bytes of the data in a single memory transaction. It comprises receiving packets of the data; generating one or more memory transactions for each of the packets of the data, wherein each memory transaction comprises a payload having a size of m bytes, and wherein the payloads contain the data; and transmitting the memory transactions to the memory; wherein m=kn and k is an integer.

Particular implementations can include one or more of the following features. k=1. Implementations comprise directing each one of the memory transactions corresponding to one of the packets to a different one of the memory banks than the one of the memory banks to which the previous one of the memory transactions corresponding to the one of the packets was directed. Implementations comprise directing each one of the memory transactions corresponding to one of the packets to the same buffer of the memory until all of the memory banks at that buffer contain data from the one of the packets.

In general, in one aspect, the invention features a method and computer program for transmitting data from a memory agent to a memory comprising a plurality of memory banks, wherein the memory can receive no more than n bytes of the data in a single memory transaction. It comprises receiving data from the memory agent; generating one or more of the memory transactions, wherein each of the memory transactions comprises a payload having a size of m bytes, and wherein the payloads contain the data; and transmitting the memory transactions to the memory; wherein m=kn and k is an integer.

Particular implementations can include one or more of the following features. k=1. Implementations comprise directing each one of the memory transactions to a different one of the memory banks than the one of the memory banks to which the previous one of the memory transactions was directed.

The details of one or more implementations are set forth in the accompanying drawings and the description below. Other features will be apparent from the description and drawings, and from the claims.

DESCRIPTION OF DRAWINGS

FIG. 1 is a block diagram of a conventional dynamic random access memory (DRAM).

FIG. 2 shows a memory system according to a preferred embodiment.

FIG. 3 is a flowchart for a process performed by the memory controller of FIG. 2 according to a preferred embodiment.

FIG. 4 shows a packet switch connected to a network such as the Internet according to a preferred embodiment.

FIG. 5 is a flowchart for a process performed by the switch of FIG. 4 according to a preferred embodiment.

FIG. 6 depicts the memory array of a DRAM such as the memory array of the DRAM of FIG. 1.

The leading digit(s) of each reference numeral used in this specification indicates the number of the drawing in which the reference numeral first appears.

DETAILED DESCRIPTION

FIG. 1 is a block diagram of a conventional dynamic random access memory (DRAM) 100. DRAM 100 comprises a memory array 102 and a DRAM controller 104. Memory array 102 comprises a plurality of memory cells (not shown) each capable of storing one or more bits of data. The memory cells are arranged into multiple memory banks that are selected by DRAM controller 104 according to a “bank select” control signal. For example, the memory cells can be arranged into four memory banks, shown in FIG. 1 as bank 0 through bank 3, according to a two-bit bank select signal.

One well-known disadvantage of using DRAM is that each access to a memory bank (for example, writing data to the memory bank) requires certain house-keeping tasks be performed before the next access to that memory bank. Therefore two consecutive accesses to different memory banks require less time than two consecutive accesses to the same memory bank. Embodiments of the invention utilize this property of DRAM to increase the overall performance of the DRAM, as described in detail below.

FIG. 2 shows a memory system 200 according to a preferred embodiment. Memory system 200 comprises a memory controller 202 that provides a memory agent 206 with access to a memory 204 such as DRAM 100 of FIG. 1. Memory controller 202 comprises an agent interface 208 for communicating with memory agent 206, a memory interface 210 for communicating with memory 204, and a memory controller core 212 for passing data between agent interface 208 and memory interface 210. Memory controller 202 can be fabricated using conventional electronic digital logic devices. Agent 206 can be a conventional processor or some other sort of electronic device.

FIG. 3 is a flowchart for a process 300 performed by memory controller 202 according to a preferred embodiment. Agent interface 208 of memory controller 202 receives data from memory agent 206 (step 302). Memory controller core 212 generates one or more memory transactions (step 304). Each memory transaction preferably comprises a command and a payload containing some or all of the data received from memory agent 206. Memory interface 210 then transmits the memory transactions to memory 204 (step 306).

Another important property of a DRAM is its burst size, which is the maximum number of bytes n of data that the DRAM can accept in a single memory transaction. In a preferred embodiment, memory controller core 212 fixes the size of the payload of each memory transaction at m bytes, where m=kn and k is an integer. In one embodiment, k=1. This process increases the probability that each memory transaction will access a different memory bank than the previous memory transaction, thereby increasing the performance of memory 204.

In one embodiment, memory controller core 212 ensures that each memory transaction is directed to a different one of the memory banks than the previous memory transaction. This technique is preferably implemented as follows. For each memory transaction, memory controller 202 provides a memory address over a memory address bus to memory 204. One or more lines of the memory bus are supplied to the bank select input of memory 204. For example, assume memory 204 uses 32-bit memory addresses so that, for each memory transaction, memory controller 202 provides a 32-bit memory address [31:0] to memory 204. Further assume that the burst size of memory 204 is 128 bytes. Therefore, according to a preferred embodiment, memory controller core 212 ensures that each memory transaction has a 128-byte payload. Because 128=2⁷, bits 7 and 8 of the address [31:0] bus are connected to the two bank select inputs of memory 204. Therefore consecutive memory transactions are directed to the memory banks in order 0-1-2-3-0-1-2-3 and so on. Of course, it is not necessary to proceed in this order, so long as no memory bank is accessed by two consecutive memory transactions.

Embodiments of the invention are particularly useful in network packet switches, but are not limited to switches. For example, embodiments of the invention are also useful in traffic managers, a multi-port media access controllers (MAC), and the like. FIG. 4 shows a packet switch 402 connected to a network 406 such as the Internet according to a preferred embodiment. Packet switch 402 comprises a memory 404 such as DRAM 100 of FIG. 1 and a plurality of ports 408A through 408N that communicate with network 406. Switch 402 also comprises a memory interface 410 for communicating with memory 404, and a memory controller core 412 for passing data between ports 408 and memory interface 410. Memory controller core 412 can be fabricated using conventional electronic digital logic devices. Ports 406 and memory interface 410 can be fabricated according to techniques well-known in the relevant arts.

FIG. 5 is a flowchart for a process 500 performed by switch 402 according to a preferred embodiment. A port 408 receives a packet of data from network 406 (step 502). Memory controller core 412 generates one or more memory transactions (step 504). Each memory transaction comprises a command and a payload containing some or all of the data in the packet. Memory interface 410 then transmits the memory transactions to memory 404 (step 506).

As discussed above, an important property of a DRAM is its burst size, which is the maximum number of bytes n of data that the DRAM can accept in a single memory transaction. In a preferred embodiment, memory controller core 412 fixes the size of the payload of each memory transaction at m bytes, where m=kn and k is an integer. In one embodiment, k=1. This process increases the probability that each memory transaction will access a different memory bank than the previous memory transaction, thereby increasing the performance of memory 404.

Of course, the number of bytes of data b in a received packet may not be evenly divisible by the number of bytes of data in the payload of each memory transaction; that is, the remainder of b/m may not be zero. Some DRAMs possess a feature commonly referred to as “interrupted burst,” which allows a memory transaction to have a payload that is less than m bytes. If memory 404 does not have the interrupted burst feature, then the payload of the final memory transaction sent to memory 404 for a given packet can be padded with pad data (that is, all ones, all zeroes, or some other bit pattern). In such embodiments, memory controller core 412 tracks the pad so that it is not transmitted when the packet is retrieved from memory for transmission to network 406. For example, memory controller core 412 can record the length of the pad, and then remove the pad when assembling the packet for transmission.

On the other hand, if memory 404 supports interrupted burst, then the payload of the final memory transaction sent to memory 404 for a given packet can be reduced from m bytes. In such embodiments, memory controller core 412 writes the first memory transaction to the first address in the next memory bank to ensure that the payload of each memory transaction is stored in a single memory bank.

In one embodiment, memory controller core 412 ensures that each memory transaction is directed to a different one of the memory banks than the previous memory transaction. This technique is preferably implemented as follows. For each memory transaction, switch 402 provides a memory address over a memory address bus to memory 404. one or more lines of the memory bus are supplied to the bank select input of memory 404. For example, assume memory 404 uses 32-bit memory addresses so that, for each memory transaction, memory controller core 412 provides a 32-bit memory address to memory 404. Further assume that the burst size of memory 404 is 128 bytes. Therefore, according to a preferred embodiment, memory controller core ensures that each memory transaction has a 128-byte payload. Because 128=2⁷, bits 7 and 8 of the address bus are connected to the two bank select inputs of memory 404. Therefore consecutive memory transactions are directed to the memory banks in order 0-1-2-3-0-1-2-3 and so on. Of course, it is not necessary to proceed in this order, so long as no memory bank is accessed by two consecutive memory transactions.

In many conventional packet switches, the memory is divided into equally-sized memory. Each block comprises one or more buffers and is the size of the maximum size of the packets handled by the switch. To ensure that a packet does not overrun a block of memory, conventional switches simply always write the first memory transaction of a packet to the first buffer of the block in the first memory bank. However, in embodiments of the present invention, while the first memory transaction of a packet is written to the first buffer of a memory block, it is often written to a memory bank other than the first memory bank. Therefore, to ensure that a packet does not overrun a memory block, memory controller core 412 ensures that each one of the memory transactions corresponding to a packet is written to the same buffer of the memory until all of the memory banks in that buffer contain data from the packet.

An example is useful to explain this technique. FIG. 6 depicts the memory array 600 of a DRAM such as memory array 102 of DRAM 100 of FIG. 1. Each row of memory array 600 represents one memory buffer, with buffer addresses increasing from the top to the bottom of the page. Each column of memory array 600 represents one memory bank. Each cell of memory array 600 is formed by the intersection of one column and one row, and is capable of storing one byte of data. In the depicted example, the maximum packet length is 12 bytes. Therefore, memory array 600 has been divided into three memory blocks B1, B2, and B3, each comprising 12 memory cells.

Referring to FIG. 6, a packet P1 has been stored in block B1 of memory array 600. Packet P1 comprises seven bytes of data, and so has been broken into seven memory transactions, each having a payload of one byte of data. In accordance with embodiments of the invention described above, the payloads of the memory transactions have been stored in memory array 600 in memory bank order 0-1-2-3-0-1-2 as bytes P1A, P1B, P1C, P1D, P1E, P1F and P1G, respectively.

Then a second packet arrives. The second packet has the maximum packet length of 12 bytes. Therefore the second packet is broken into 12 memory transactions. In accordance with embodiments of the invention described above, the payload of the first memory transaction (which generally contains the first byte of data of the packet) is written to the memory bank following the memory bank to which the last byte of the previous packet was written. In particular, referring to FIG. 6, the first byte P2A of the second packet is written to memory bank 3 in the first buffer of block B2.

According to embodiments of the present invention, the next byte P2B of the second packet is written to memory bank 0. However, if byte P2B is written to the second buffer of memory block B2, and so on, the last three bytes of the second packet will overrun memory block B2, and be written to the first buffer of the third memory block, while three cells in the first buffer of memory block B2 will remain unused.

Therefore, according to a preferred embodiment, the payloads P2B, P2C, and P2D of the next three memory transactions are written to

banks

0, 1, and 2, respectively, at the first buffer of memory block B2, so that buffer is full. Then the payload P2E of the next memory transaction is written to memory bank 3 at the next buffer. Then, the payloads P2F, P2G, and P2H of the next three memory transactions are written to

banks

0, 1, and 2, respectively, at the second buffer of memory block B2, and so on, as shown in FIG. 6, so that the second packet completely fills memory block B2, with no wasted memory cells in memory block B2, and no overrun into memory block B3, while preserving the memory bank order of access to the memory banks (here, 0-1-2-3-0-1-2-3). Of course, these embodiments of the invention can be used regardless of whether memory 404 supports interrupted burst, as will be apparent to one skilled in the relevant arts after reading this description.

Memory controller core

412 can use many techniques to keep track of the order of the packets. For example, for each packet, memory controller core 412 can record, with the packet ID, the buffer and the memory bank where the payload of the first memory transaction for that packet is stored. Then, when the packet is retrieved from memory, the same process used to write the data to memory 404 can be used to retrieve the data, thereby restoring the packet correctly.

The invention can be implemented in digital electronic circuitry, or in computer hardware, firmware, software, or in combinations of them. Apparatus of the invention can be implemented in a computer program product tangibly embodied in a machine-readable storage device for execution by a programmable processor; and method steps of the invention can be performed by a programmable processor executing a program of instructions to perform functions of the invention by operating on input data and generating output. The invention can be implemented advantageously in one or more computer programs that are executable on a programmable system including at least one programmable processor coupled to receive data and instructions from, and to transmit data and instructions to, a data storage system, at least one input device, and at least one output device. Each computer program can be implemented in a high-level procedural or object-oriented programming language, or in assembly or machine language if desired; and in any case, the language can be a compiled or interpreted language. Suitable processors include, by way of example, both general and special purpose microprocessors. Generally, a processor will receive instructions and data from a read-only memory and/or a random access memory. Generally, a computer will include one or more mass storage devices for storing data files; such devices include magnetic disks, such as internal hard disks and removable disks; magneto-optical disks; and optical disks. Storage devices suitable for tangibly embodying computer program instructions and data include all forms of non-volatile memory, including by way of example semiconductor memory devices, such as EPROM, EEPROM, and flash memory devices; magnetic disks such as internal hard disks and removable disks; magneto-optical disks; and CD-ROM disks. Any of the foregoing can be supplemented by, or incorporated in, ASICs (application-specific integrated circuits).

A number of implementations of the invention have been described. Nevertheless, it will be understood that various modifications may be made without departing from the spirit and scope of the invention. Please list any additional modifications or variations. Accordingly, other implementations are within the scope of the following claims.

Claims

1. A memory controller comprising:

at least one agent interface, wherein each agent interface is adapted to receive data from a respective memory agent;

a memory controller core adapted to generate one or more memory transactions, wherein each memory transaction comprises a payload having a size of m bytes, and wherein the payloads contain the data; and

a memory interface adapted to transmit the memory transactions to a memory, wherein the memory comprises a plurality of memory banks and can receive no more than n bytes of data in a single memory transaction;

wherein m=kn, k is an integer, and m is not equal to n.

2. The memory controller of claim 1, wherein k=1.

3. The memory controller of claim 1, wherein the memory controller core is further adapted to direct each one of the memory transactions to a different one of the memory banks than the one of the memory banks to which the memory controller core directed the previous one of the memory transactions.

4. A network switch comprising the memory controller of claim 1.

5. A traffic manager comprising the memory controller of claim 1.

6. A multi-port media access controller comprising the memory controller of claim 1.

7. An apparatus comprising:

a plurality of port means for receiving packets of data;

memory controller core means for generating one or more memory transactions for each of the packets of data, wherein each memory transaction comprises a payload having a size of m bytes, and wherein the payloads contain the data;

memory means comprising a plurality of memory bank means for storing the data, wherein the memory means can receive no more than n bytes of data in a single memory transaction; and

memory interface means for transmitting the memory transactions to the memory means;

wherein m=kn, k is an integer, and m is not equal to n.

8. The apparatus of claim 7, wherein k=1.

9. The apparatus of claim 7, wherein the memory controller core means directs each one of the memory transactions corresponding to one of the packets to a different one of the memory bank means than the one of the memory bank means to which the memory controller core means directed the previous one of the memory transactions corresponding to the one of the packets.

10. The apparatus of claim 7, wherein the memory bank means define a plurality of buffer means, and wherein the memory controller core means directs each one of the memory transactions corresponding to one of the packets to the same buffer means of the memory means until all of the memory banks at that buffer means contain data from the one of the packets.

11. A network switch comprising the apparatus of claim 7.

12. A traffic manager comprising the apparatus of claim 7.

13. A multi-port media access controller comprising the apparatus of claim 7.

14. A memory controller comprising:

at least one agent interface means each for receiving data from a respective memory agent;

memory controller core means for generating one or more memory transactions, wherein each memory transaction comprises a payload having a size of m bytes, and wherein the payloads contain the data; and

memory interface means for transmitting the memory transactions to a memory, wherein the memory comprises a plurality of memory banks and can receive no more than n bytes of data in a single memory transaction;

wherein m=kn, k is an integer, and m is not equal to n.

15. The memory controller of claim 14, wherein k=1.

16. The memory controller of claim 14, wherein the memory controller core means directs each one of the memory transactions to a different one of the memory banks than the one of the memory banks to which the memory controller core means directed the previous one of the memory transactions.

17. A network switch comprising the memory controller of claim 14.

18. A traffic manager comprising the memory controller of claim 14.

19. A multi-port media access controller comprising the memory controller of claim 14.

20. A method for transmitting data to a memory

comprising a plurality of memory banks, wherein the memory can receive no more than n bytes of the data in a single memory transaction, the method comprising: receiving packets of the data;

generating one or more memory transactions for each of the packets of the data, wherein each memory transaction comprises a payload having a size of m bytes, and wherein the payloads contain the data; and

transmitting the memory transactions to the memory;

wherein m=kn, k is an integer, and m is not equal to n.

21. The method of claim 20, wherein k=1.

22. The method of claim 20, further comprising:

directing each one of the memory transactions corresponding to one of the packets to a different one of the memory banks than the one of the memory banks to which the previous one of the memory transactions corresponding to the one of the packets was directed.

23. The method of claim 20, further comprising:

directing each one of the memory transactions corresponding to one of the packets to the same buffer of the memory until all of the memory banks at that buffer contain data from the one of the packets.

24. A method for transmitting data from a memory agent to a memory comprising a plurality of memory banks, wherein the memory can receive no more than n bytes of the data in a single memory transaction, the method comprising:

receiving packets of the data from the memory agent;

generating one or more of the memory transactions, wherein each of the memory transactions comprises a payload having a size of m bytes, and wherein the payloads contain the data; and

transmitting the memory transactions to the memory; wherein m=kn, k is an integer, and m is not equal to n.

25. The method of claim 24, wherein k=1.

26. The method of claim 24, further comprising:

directing each one of the memory transactions to a different one of the memory banks than the one of the memory banks to which the previous one of the memory transactions was directed.

27. A computer program embodying instructions executable by a computer to perform a method for transmitting data to a memory comprising a plurality of memory banks, wherein the memory can receive no more than n bytes of the data in a single memory transaction, the method comprising: receiving packets of the data;

transmitting the memory transactions to the memory;

wherein m=kn, k is an integer, and m is not equal to n.

28. The computer program of claim 27, wherein k=1.

29. The computer program of claim 27, wherein the method further comprises:

30. The computer program of claim 27, wherein the method further comprises:

31. A computer program embodying instructions executable by a computer to perform a method for transmitting data from a memory agent to a memory comprising a plurality of memory banks, wherein the memory can receive no more than n bytes of the data in a single memory transaction, the method comprising:

receiving packets of the data from the memory agent;

generating one or more memory transactions, wherein each of the memory transactions comprises a payload having a size of m bytes, and wherein the payloads contain the data; and

transmitting the memory transactions to the memory;

wherein m=kn, k is an integer, and m is not equal to n.

32. The computer program of claim 31, wherein k=1.

33. The computer program of claim 31, wherein the method further comprises:

34. An apparatus comprising:

a plurality of ports each adapted to receive packets of data;

a memory core adapted to generate one or more memory transactions for each of the packets of data, wherein each memory transaction comprises a payload having a size of m bytes, and wherein the payloads contain the data;

a memory comprising a plurality of memory banks adapted to store the data, wherein the memory can receive no more than n bytes of data in a single memory transaction; and

a memory interface adapted to transmit the memory transactions to the memory;

wherein m=kn, m is not equal to n, and k is an integer.

35. The apparatus of claim 34, wherein k=1.

36. The apparatus of claim 34, wherein the memory controller core is further adapted to direct each one of the memory transactions corresponding to one of the packets to a different one of the memory banks than the one of the memory banks to which the memory controller core directed the previous one of the memory transactions corresponding to the one of the packets.

37. The apparatus of claim 34, wherein the memory banks define a plurality of buffers, and wherein the memory controller core is further adapted to direct each one of the memory transactions corresponding to one of the packets to the same buffer of the memory until all of the memory banks at that buffer contain data from the one of the packets.

38. A network switch comprising the apparatus of claim 34.

39. A traffic manager comprising the apparatus of claim 34.

40. A multi-port media access controller comprising the apparatus of claim 34.